Aiming at the problem of unsupervised cross-domain migration for person re-identification,an algorithm based on domain discriminative network and doma in adaptation is proposed.Firstly,the improved ResNet-50training supervised domain discrimina tive network model is used,and the shared space component is added to obtain the feature invariant attribute,which is used for the inter-class classification image,and the classification perfor mance of the model is improved based on the contrast loss and the difference loss.Secondly,the do main invariant migration method is used to derive the feature invariant attribute from the sour ce domain dataset and should be on the unmarked target domain dataset.Finally,the matching query image and the gallery image in the shared space perform cross-domain person re-identificatio n.In order to verify the validity of the algorithm,experiments were carried out on the CUHK03,Market-1501and DukeMTMC-reID datasets.The accuracy of the algorithm in Rank-1reached 34.1%,38.1% and 28.3%, respectively,and reached 34.2%,17.1% and 17.5% in mAP.Finally,the necessity of each component of the model in the training phase is verified.The results show that the perfor mance of the proposed algorithm on large-scale datasets is better than some existing unsuper vised person re-identification methods,even close to The performance of some tr aditional supervised learning methods.